Hi everyone!
\nI have just published a new notebook on using vegetation indices for RGB images to segmentate Cassava leaves. While there are many research papers in remote sensing and agriculture that use these image processing techniques, I do not know of any paper that successfully use them as a preprocessing step to improve a deep learning model's performance.
\nThese types of preprocessing methods are quite common and I seem to recall them being key to the best performing models in competitions like APTOS 2019. I have only played a little with the preprocessing methods I am presenting for this competition and the results are neutral, so far.
\nTherefore, I would be interested in hearing people's opinion on the potential success or failure of such approach for this competition. Are we already reaching a cap in terms of learning from visual information?
\nAlso, I would be happy to work on this competition with a team. I might have a bit more domain knowledge to share… So feel free to get in touch!
", "messages": [ { "id": 1122742, "postDate": "2020-12-22T17:12:27.533Z", "content": "Hi everyone!
\nI have just published a new notebook on using vegetation indices for RGB images to segmentate Cassava leaves. While there are many research papers in remote sensing and agriculture that use these image processing techniques, I do not know of any paper that successfully use them as a preprocessing step to improve a deep learning model's performance.
\nThese types of preprocessing methods are quite common and I seem to recall them being key to the best performing models in competitions like APTOS 2019. I have only played a little with the preprocessing methods I am presenting for this competition and the results are neutral, so far.
\nTherefore, I would be interested in hearing people's opinion on the potential success or failure of such approach for this competition. Are we already reaching a cap in terms of learning from visual information?
\nAlso, I would be happy to work on this competition with a team. I might have a bit more domain knowledge to share… So feel free to get in touch!
", "rawMarkdown": "Hi everyone!\n\nI have just published a [new notebook](https://www.kaggle.com/frlemarchand/cassava-leaf-segmentation-using-vegetation-indices) on using vegetation indices for RGB images to segmentate Cassava leaves. While there are many research papers in remote sensing and agriculture that use these image processing techniques, I do not know of any paper that successfully use them as a preprocessing step to improve a deep learning model's performance.\n\nThese types of preprocessing methods are quite common and I seem to recall them being key to the best performing models in competitions like [APTOS 2019](https://www.kaggle.com/c/aptos2019-blindness-detection/overview). I have only played a little with the preprocessing methods I am presenting for this competition and the results are neutral, so far.\n\nTherefore, I would be interested in hearing people's opinion on the potential success or failure of such approach for this competition. Are we already reaching a cap in terms of learning from visual information?\n\nAlso, I would be happy to work on this competition with a team. I might have a bit more domain knowledge to share... So feel free to get in touch!", "votes": 5 } ], "comments": [], "raw_markdown_by_id": { "1122742": "Hi everyone!\n\nI have just published a [new notebook](https://www.kaggle.com/frlemarchand/cassava-leaf-segmentation-using-vegetation-indices) on using vegetation indices for RGB images to segmentate Cassava leaves. While there are many research papers in remote sensing and agriculture that use these image processing techniques, I do not know of any paper that successfully use them as a preprocessing step to improve a deep learning model's performance.\n\nThese types of preprocessing methods are quite common and I seem to recall them being key to the best performing models in competitions like [APTOS 2019](https://www.kaggle.com/c/aptos2019-blindness-detection/overview). I have only played a little with the preprocessing methods I am presenting for this competition and the results are neutral, so far.\n\nTherefore, I would be interested in hearing people's opinion on the potential success or failure of such approach for this competition. Are we already reaching a cap in terms of learning from visual information?\n\nAlso, I would be happy to work on this competition with a team. I might have a bit more domain knowledge to share... So feel free to get in touch!" } }